GemmaWithLoRA: A New Approach to Click Fraud Detection

With the rapid development of artificial intelligence technology, pre-trained large language models (LLMs) have demonstrated their powerful application potential in various fields. Click fraud is a serious issue in the digital advertising ecosystem, especially in the pay-per-click (PPC) model, where fraudulent activities significantly undermine the effectiveness of ads and pose security risks. While many traditional fraud detection methods have achieved relatively good accuracy, they still face poor generalization and extensive feature engineering issues. This paper explores using LLMs for click fraud detection by leveraging the powerful feature extraction capabilities and superior generalization performance of the Gemma-2b model, combined with LoRA fine-tuning technology. Extensive experiments on the TalkingData2017 dataset showed that accuracy was achieved at 83°/0, demonstrating the effectiveness of LLMs in click fraud detection tasks.